From 002e3fef38b1c2db583ff0c887eae6dfd1fd69d0 Mon Sep 17 00:00:00 2001
From: "Joseph R. Quinn" <423821+quinnjr@users.noreply.github.com>
Date: Fri, 1 May 2026 12:15:35 -0400
Subject: [PATCH] feat(users): backfill user.name from oauth_github.login

`users.name` is currently set only when the GitHub profile has a
display name, so a sizable fraction of accounts have it as NULL.
That makes the column unusable as a guaranteed identifier and means
any feature that wants a non-NULL name has to fall back to gh_login
on its own.

Fill in the NULLs with `oauth_github.login` (the GitHub username)
so every account has a name. The DO-block walks `users.id` in
5000-row windows and COMMITs between batches so row locks release
between iterations and concurrent session-login upserts aren't
blocked. lock_timeout / statement_timeout cap each batch in case
of unexpected contention.

Idempotent: filtered on `users.name IS NULL`, so re-running is a
no-op once every account has a name. Follows the same out-of-band
pattern as data_oauth_github.sql -- run with `psql -f <file>` and
do NOT wrap in BEGIN/COMMIT, since the COMMIT inside the DO block
requires the block to be at the top level.

The session-login upsert path is unchanged: future logins will
still write the GitHub display name when one is set, overwriting
any backfilled value with the user's actual chosen name.
---
 migrations/data_user_name_backfill.sql | 40 ++++++++++++++++++++++++++
 1 file changed, 40 insertions(+)
 create mode 100644 migrations/data_user_name_backfill.sql

diff --git a/migrations/data_user_name_backfill.sql b/migrations/data_user_name_backfill.sql
new file mode 100644
index 00000000000..dca63de7721
--- /dev/null
+++ b/migrations/data_user_name_backfill.sql
@@ -0,0 +1,40 @@
+-- Backfill `users.name` from `oauth_github.login` for users whose `name` is
+-- still NULL. `users.name` is currently set only when the GitHub profile has
+-- a display name, so users without one have NULL there; this fills those in
+-- with the GitHub username so every account has a non-NULL name.
+--
+-- Iterates by `users.id` range and COMMITs between batches so row locks
+-- release between iterations and concurrent traffic isn't blocked. The DO
+-- block must be run outside an explicit transaction (`psql -f <file>` is
+-- fine; do NOT wrap with BEGIN/COMMIT, since COMMIT inside DO requires the
+-- block to be at the top level).
+--
+-- Idempotent: re-running the file is a no-op once every account has a name,
+-- because the UPDATE filters on `users.name IS NULL`.
+
+SET lock_timeout = '5s';
+SET statement_timeout = '60s';
+
+DO $$
+DECLARE
+    lo INT;
+    hi INT;
+    pos INT;
+    batch_size CONSTANT INT := 5000;
+BEGIN
+    SELECT MIN(id), MAX(id) INTO lo, hi FROM users WHERE name IS NULL;
+    IF lo IS NULL THEN RETURN; END IF;
+
+    pos := lo;
+    WHILE pos <= hi LOOP
+        UPDATE users
+        SET name = oauth_github.login
+        FROM oauth_github
+        WHERE oauth_github.user_id = users.id
+          AND users.name IS NULL
+          AND users.id >= pos
+          AND users.id < pos + batch_size;
+        COMMIT;
+        pos := pos + batch_size;
+    END LOOP;
+END $$;